Fitter function for a probabilistic index model

Share:

Description

This is the basic computing engine called by pim to get the estimates for the coefficients and the variance- covariance matrices. This function currently only spits out these components using the sandwich estimators.

Usage

1
2
pim.fit(x, y, link = "logit", estim = "estimator.nleqslv", start = rep(0,
  ncol(x)), vcov.estim = "sandwich.vcov", weights = NULL, penv, ...)

Arguments

x

a model matrix with as many rows as y.

y

a vector with the pseudo-responses

link

a character vector with a link function

estim

a character vector or a function indicating the solver to be used for estimating the coefficients. By default this is the function nleqslv. Other possibilities are given in the help page on estimators.

start

A numeric vector with the starting values for the fitting algorithm, if required.

vcov.estim

a function to determine the variance-covariance matrix. Possibilities are sandwich.vcov and link{score.vcov}. Defaults to sandwich.vcov

weights

currently not implemented

penv

An environment, pim.environment or pim.poset object containing the poset functions. Alternatively this can be a list of two numeric vectors, containing the poset indices for the left and right side of the pim.

...

Further arguments that need to be passed to the estimation function. The most relevant is construct, allowing you to write your own score function for the numerical optimization. See also estimators

Value

A list with the following elements

coefficients

a numeric vector with the coefficients

vcov

a numeric matrix with the variance-covarianc matrix for the coefficients

fitted

a numeric vector with the raw fitted values

estim

a list with two components named coef and vcov containing information on the used estimators for both.

See Also

model.matrix for how to construct a valid model matrix for a pim, pim for the general user interface

Want to suggest features or report bugs for rdrr.io? Use the GitHub issue tracker.